DECISIVe : Workshop on Dealing with Cognitive Biases in Visualisations, IEEE VIS2014, Nov 9th 2014, Paris

Accounting for Availability Biases in Information Visualization

Evanthia Dimara 1,2 Pierre Dragicevic 1 Anastasia Bezerianos 2,1 [email protected] [email protected] [email protected]

1INRIA 2Univ Paris-Sud & CNRS (LRI)

ABSTRACT effortlessly. Schwarz et al [16] asked one group of participants to recall 12 examples of their past assertive behavior, and another The availability heuristic is a strategy that people use to make group to recall only 6. After that, they rated their assertiveness. quick decisions but often lead to systematic errors. We propose The 12 examples were harder to be recalled than the 6, so the first three ways that visualization could facilitate unbiased decision- group rated themselves as less assertive than the second group. making. First, visualizations can alter the way our memory stores Retrievability is often related to a) familiarity [20] or what the events for later recall, so as to improve users' long-term Whittlesea [23] refers to as the "illusion of pastness"; b) saliency intuitions. Second, the known biases could lead to new where one instance elicits more attention than another [18]; and c) visualization guidelines. Third, we suggest the design of decision- recency where for example the serial presentation of information making tools that are inspired by heuristics, e.g. suggesting may affect memorization [22]. intuitive approximations, rather than target to present exhaustive comparisons of all possible outcomes, or automated solutions for - Bias due to the effectiveness of the search set: The choosing decisions. generation of a search set depends also on the performed search task. When we ask to compare the instances of the word ‘love' Categories and Subject Descriptors with the word ‘door’ the first seems more frequent. A main reason H.5.2 [Information Interfaces and Presentation]: User Inter- for this is that besides the comparison of words, there is a hidden faces - Graphical user interfaces task of recalling contexts in which these words appear. It is generally easier to recall abstract contexts than concrete ones [8]. General Terms - Bias of imaginability: When the frequency of an instance is Design, Human Factors not stored in memory, we sometimes generate this frequency according to some rule. For example when we want to estimate which is more frequent, the existence of committees of 8 members

Keywords or of 2, we will mentally construct committees and rate them by Availability Heuristic, Cognitive Bias, Visualizations, Decision- the ease of this construction. The mental construction of 2 Making member committees is easier, and thus may be considered as most frequent [20]. In real life imaginability biases can lead us to 1. INTRODUCTION overestimate some risks with vivid scenarios and underestimate We all want to make good decisions. However, decision- dangerous risks that are hard to conceive. making judgments often involve approximate estimations of - Bias due to illusory correlation: When two events co-occur probabilities and frequencies. In order to reduce the complexity of people tend to overestimate the frequency of natural association. estimation people rely on a limited number of strategies. One of For example it is common to patients with paranoia to have these strategies is called Availability Heuristic [14]. peculiar eyes. This association misled undergraduate clinicians to The availability heuristic is a rule of thumb in which decision diagnose as paranoid patients with no other symptoms related to makers “estimate the frequency or probability by the ease with paranoia in their medical data, simply because they were guided which instances or associations could be brought to mind”[20]. by a given picture of the patient with peculiar eyes [4][5]. For example, news about a terrible plane crash may temporarily Availability bias affects the decision-making ability in an alter our feelings on flight safety. This heuristic simplifies some unconscious way, and can lead people to irrational decisions. We otherwise very difficult judgments, and it is usually effective since believe we can better support decision-making, through the design in principle it is easier to recall or imagine common events than of visualizations that take into account these factors that influence uncommon ones. decision-making. We illustrate this in a voting scenario that we However, apart from the actual frequency or probability, imagine takes place with and without hypothetical visualizations other factors affect the ease of recalling instances, and thus designed to account for biases. estimating frequencies for making decisions. Some of these factors affecting recall, illustrated by Tversky and Kahneman 2. THE VOTING DECISION [21], often lead to systematic errors: Imagine that one needs to decide which political candidate to - Bias due to retrievability of instances: People evaluate the vote for. There are three steps. As a first step, she shapes an probability of an event as higher, when they retrieve its instances opinion on which are the important personal and society issues

Konstanzer Online-Publikations-System (KOPS) URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-0-329436 based on her past exposure to information (media, social [24]. Thus visualizations can reinforce the importance of some environment, personal experiences etc.). Second, she investigates information due to some known biases (e.g., presenting them last). the candidates’ former actions, background and current positions, and estimates their ability and willingness to solve the important Moreover, if the magnitude of a visual variable does not reflect its society issues. Finally, she compares all the alternatives and real impact, we may reinforce not only visual perception bias (bigger is more important), but also retrievability biases (bigger decides on a candidate. may be easier to remember). For example, consider the two In an ideal world, voters are aware of their position in the political candidates suggest either the increase of unemployment complex political landscape, understand statistical analysis data allowance, or the tax exemption of families with a lot of children. and micro-macro economics, and have endless memory capacity Both are fair measures, but unemployment applies to a larger part and time to process all the relative candidates’ history. In reality, of the population. However, families with many children may not voters usually simplify this decision using heuristics. However, as be able to survive with the current tax policy. A visualization we discussed, the common heuristics like availability may lead where the visual variable depends exclusively on the population one to pick a candidate according to, for example: meaningless size is legible, but may lead voters to evaluate the policies only actions that media over-cover, without important impact in the according to the population criterion, even for voters who have 10 society [6]; the sequence of their presentation in the public debate children themselves. Visualizations, in addition to what the media [12]; the vividness of the way they talk [15][13][2] or even can offer, should be able to also display a customized perspective whether their victory is an event easy to envision [3]. and alternative views of the data. These customized views of the data will be the ones most likely retrieved from their memory 3. INFOVIS ON A COMPLEX DECISION during the decision process. The visualization design should also take into account the The actual challenge on the three decision-making steps in biases due to imaginability. When it comes to radical ideas, the the voting problem is how to filter, understand, recall and mind’s inability to construct the outcome of this idea can lead a compare information. In principle, this challenge is related to the person to consider it as impossible to happen. In the voting infovis objectives. But how could visualizations actually assist a problem, if a candidate proposes “decentralization of state power voter to reduce availability biases? to local communities”, the voters may reject it not only because Visualizations to aid recall: Let’s think of a scenario with simple they disagree, but also because the outcome is an event hard to hypothetical visualizations involved in the voting problem. envision. A conceptualized, but vivid, map representation of the Consider also two alternative policies: a consolidation of the idea of decentralization could alter the voter’s willingness to national health system focusing on cancer cure, and a high-cost accept a change. terrorism counteraction. People tend to make wrong estimations Visualizations inspired by heuristics: We mostly discussed so far on most probable causes of death in their country [7], misled by how to present the information to lead to effective and unbiased media. In contrast, imagine the voter had access to a map with decisions. However, the decision-making process itself can be stacked (men/women) bar charts of death causes (society issues hard even when all the information we need is available. awareness) that she can filter. For the simplicity of the argument Automated decision-making tools that give explicit answers here we assume a voter who wants to maximize her own self- according to probability computations, may not always feel interest. Thus, the displayed causes can be filtered according to intuitive and understandable even by experts [9]. They often her family’s medical records and other individual characteristics restrict users by expecting a very particular input, or ignoring (personal issues awareness). The user interacts with the other context-relevant information that the users may have [19]. visualization, composing a view that is focused on her interests. On the other hand common visualization tools often hide From the amount of information that she was able to process, she uncertainty in the data [17] [1] and do not actually shield decision captures a snapshot of her self-constructed view of the makers against perceptual and cognitive biases [10][14][11] visualization to save among her personal notes. In the last step of Visualization tools are currently designed and evaluated based on the decision-making, she evaluates the policies of each candidate. data retrieval and insight tasks, rather than on the ultimate and Based on her memorable interaction experience with the crucial task of decision-making [1]. Thus we could drive some visualization, she intuitively evaluates which election promise has inspiration of how heuristic strategies like availability, simplify greater impact in her life. decision tasks, find the effective tradeoff among simplicity and Visualizations could thus educate decision makers to develop accuracy, and apply this analogy on new visualization tools. That unbiased intuitions of their surroundings. However, as we saw in is, decision tools may need to allow some imperfection for the the previous example, this does not only imply visual comparisons sake of understandability. of choices and consequences. The availability heuristic succeeds when memory stores the frequent events in an easy-to-recall way. 4. CONCLUSION In our example, the visualization facilitates the user’s memory by capturing her self-constructed summary. Thus, visualizations could go beyond simply showing all possible alternatives. We discussed so far how information visualization could Visualizations should also help decision makers easily recall the eliminate biases due to availability, when we visualize the actual important take-away information. probabilities, rearrange the sequence of information, increase their saliency or filter a subset of them for later recall. While designing Visualizations to remove biases: So visualizations can help deal such systems, our goal should not be to eliminate the use of with biases. But visualization designers can also reuse the heuristics that can cause these biases altogether, but rather to knowledge of robust biases already studied in psychology exploit them helping users make better decisions. literature to make better visualizations. Studies on how the candidates’ order in the set of ballot papers affects the vote rank, To sum up, we suggest that information visualization can reduce confirm that “recency effect” [22] occurs also in visualizations availability biases and assist decision-making in three ways. First, we can take advantage of the good use of availability heuristics measurements. Journal of Abnormal and Social Psychology, and improve users' long-term intuitions. Second, visualizations 59, 1-9. could provide design techniques that eliminate the known [13] Nisbett, R. R., & Ross, L. 1980. Human inference: Strategies availability biases. Third, we can investigate new decision-making and shortcomings of social judgment. Englewood Cliffs, NJ: tools that target to inspired by rather than replace the way that Prentice-Hall availability heuristics work. [14] Plous, S. (1993). The psychology of judgment and decision- making. Mcgraw-Hill Book Company. [15] Reyes, R. M., Thompson, W. C., & Bower, G. H. 1980. Judgmental biases resulting from differing availabilities of 5. REFERENCES arguments. Journal of Personality and Social Psychology, 39, 2-12. [1] Amar, R. A. , Stasko J. T. , 2005. Knowledge Precepts for Design and Evaluation of Information Visualization, IEEE [16] Schwarz, N., Bless, H., & Strack, F. (1991). Ease of Transactions on Visualization and Computer Graphics - Retrieval as Information: Another Look at the Availability TVCG, vol. 11, no. 4, pp. 432-442, Heuristic. Journal of Personality and Social Psychology, 61 (2), 195-202. [2] Borgida, E., & Nisbett, R. E. 1977. The differential impact of abstract vs. concrete information on decisions. Journal of [17] Streit A, Pham B., Brown R. A Spreadsheet Approach to Applied Social Psychology, 7, 258-271 Facilitate Visualization of Uncertainty in Information IEEE Transactions on Visualization and Computer Graphics - [3] Carroll, J.S. 1978. The effect of imagining an event on TVCG, vol. 14, no. 1, pp. 61-72, 2008 expectations for the event: An interpretation in terms of availability heuristic. Journal of Experimental Psychology, [18] Taylor, S.E., & Fiske, S. T. 1978. Salience, attention and 14, 88-96. attribution: Top of the head phenomena. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 11, [4] Chapman, L. J., & Chapman, J. P. 1967. Genesis of popular pp. 249-288). New York: Academic Press but erroneous psychodiagnostic observations. Journal of Abnormal Psychology, 73, 193-204 [19] Teevan J., Alvarado C., Ackerman M. S., Karger D. R. The Perfect Search Engine Is Not Enough: A Study of [5] Chapman, L. J., & Chapman, J. P. 1969. Illusory Correlation Orienteering Behavior in Directed Search Computer Human as an Obstacle to the Use of Valid Psychodiagnostic Signs. Interaction - CHI, pp. 415-422, 2004 Journal of Abnormal Psychology, 74, 271-80 [20] Tversky, A., & Kahneman, D. 1973. Availability: A heuristic [6] Combs, B., & Slovic, P. 1979. Newspaper coverage of for judging frequency and probability. Cognitive Psychology, causes of death. Journalism Quarterly, 56, 837 - 843, 849 5, 207-232 [7] “Death Odds.” (1990, September 24). Newsweek, p. 10. [21] Tversky, A., & Kahneman, D. 1974. Judgment under [8] Galbraith, R. C., & Underwood, B.J. 1973. Perceived uncertainty: Heuristics and biases. Science, 185, 1124-1130 Frequency of Concrete and Abstract Words. Memory and [22] Ward, W. C., & Jenkins, H. M. 1965. The display of Cognition 1, 56-60 information and the judgment of contingency. Canadian [9] Gersh, J. R., McKneely, J. A., & Remington, R. W. (2005). Journal of Psychology, 19, 231-241 Cognitive Engineering: Undertanding Human Interaction [23] Whittlesea, B. W. A., Jacoby, L. L., & Girard, K. 1990. with Complex Systems. Johns Hopkins APL Technical Illusions of immediate memory: Evidence of an attributional Digest, 26, 4, 377-382. basis for feelings of familiarity and perceptual quality. [10] Hullman J., Diakopoulos N., 2011. Visualization Rhetoric: Journal of Memory and Language, 29, 716–732. Framing Effects in Narrative Visualization, IEEE [24] Wood, J., Badawood, D., Dykes, J. & Slingsby, A. 2011. Transactions on Visualization and Computer Graphics - BallotMaps: Detecting name bias in alphabetically ordered TVCG, vol. 17, no. 12, pp. 2231-2240 ballot papers. IEEE Transactions on Visualization and [11] Kahneman, D. (2011). Thinking, fast and slow. Macmillan. Computer Graphics, 17(12), pp. 2384-2391. [12] Miller, N., & Campbell, D. T. 1959. Recency and primacy in persuasion as a function of the timing of speeches and